Beruflich Dokumente
Kultur Dokumente
T
his review paper is intended to serve as a touchstone on the state of the
science of calculating the flow of water through unsaturated porous me-
dia. As active researchers we believe it is good to occasionally take stock
in what has been accomplished up to date, where things may be headed, and what
important issues remain. This review concentrates on computational modeling of
flow through the unsaturated zone via Richards’ equation.
This effort can help provide some focus to the research community and serve
to help propel new research forward, particularly by those just joining the field.
There are many practical problems that require calculation of one-, two-, and three-
dimensional fluxes in the unsaturated zone. For a much broader review of modeling
soil processes, we recommend the recent paper from Vereecken et al. (2016). A
detailed review of mathematical models of infiltration can be found (Assouline,
2013), while Paniconi and Putti (2015) provide historical perspective and an
overview of modeling Richards’ equation in the context of catchment hydrology.
Core Ideas The equation attributed to Richards (1931) that describes the flow of water
• The numerical solution of Richards’ through unsaturated porous media under the action of capillarity and gravity was
equation remains challenging. first published by the English mathematician and physicist Lewis Fry Richardson
• Space/time discretization affects both in 1922 (Richardson, 1922). Therefore, the equation would rightly be called
computational effort and accuracy. “Richardsons’ equation”, although in this paper we continue to call it by its
common name, Richards’ equation.
• Adaption of space and time
discretizations produces benefits.
Soil Sci. Soc. Am. J. 81:1257–1269
• Dissemination of codes and improved
doi:10.2136/sssaj2017.02.0058
documentation are needed. Received 16 Feb. 2017.
Accepted 4 June 2017.
• Recent reformulation of one- *Corresponding author: (fogden@uwyo.edu)
dimensional Richards’ equation © Soil Science Society of America. This is an open access article distributed under the CC BY-NC-ND
shows promise. license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Soil Science Society of America Journal
Richards’ equation can be seen as a simplification of the ∂y ∂ ∂y
standard two-phase flow formulation for a gas and water phase in S s S a ( y ) + c ( y ) ∂ t − ∂ z K ( y ) ∂ z −1 =
0 [5]
a porous medium where the pressure gradient required to drive
flow of the gas phase is ignored due to the large mobility contrast where Ss [L–1] is the specific storage and Sa [-] is the saturation of
between the water and gas phases. One of the most interesting the aqueous phase (Miller et al., 1998). While ¶q/¶t = c(y)¶y/¶t
aspects of the Richards’ equation is that despite its ease of holds at the continuous level through the chain rule, the
derivation, it is arguably one of the most difficult equations to nonlinearity of c(y) makes it impossible to maintain this equality
reliably and accurately solve in all of hydrosciences. uniformly at the discrete level, and the pure head-based form in Eq.
Richards’ equation can be written in a number of forms, [4] or [5] is not generally mass conservative. There are techniques
namely the water content, mixed water content and capillary like a chord-slope approximation (Rathfelder and Abriola, 1994)
head form, and the head form. In one-dimension, the so-called that can be used in specific cases or flux updating that can promote,
“mixed water content form”, because it mixes the water content θ but not ensure, mass conservation (Kirkland et al., 1992). In
with the capillary head y(θ), is: general though, high accuracy in the time integration of Eq. [4] is
essential to avoid mass balance errors (Tocci et al., 1997), and Eq.
∂q ∂ ∂ y(q)
= K (q ) −1 [1] [3] together with expansion of the nonlinearity θ(y) in the storage
∂t ∂z ∂z
change term, ¶θ/¶t, (Allen and Murphy, 1985; Celia et al., 1990)
where z is vertical coordinate (positive downward)[L]; t is time is preferred when low-order time integration is used.
[T]; q equals q(z,t) equals volumetric soil moisture content [-]; Richards’ equation is a nonlinear, degenerate elliptic-
y(q) is empirical soil hydraulic capillary head function [L]; K(q) parabolic partial differential equation (List and Radu, 2016).
= empirical unsaturated hydraulic conductivity function [L T–1]. The head-based forms of the equation, Eq. [3–5], transition
Equation [1] is often written solely as a function of the from parabolic to elliptic as the solution domain nears saturation
water content by introducing the chain rule: unless storage effects are included (Ss > 0), in which case Eq. [5]
remains parabolic. The nonlinearity and transition of type in
∂q ∂ ∂q
= D(q) − K (q) [2] different portions of the problem domain make it very difficult to
∂t ∂z ∂z solve using traditional analytical techniques, and it is impossible
where D(θ) = K(θ)(¶y(θ)/¶θ) is referred to as the soil-water to solve in closed form except for a small number of special cases
diffusivity [L2 T–1]. As in Eq. [1], the first term in parentheses (Miller et al., 1998). This difficulty is only exacerbated by the
captures the effects of capillarity, while second term in range of initial and boundary conditions encountered in practice
parentheses represents the effect of gravity-driven flux. as well as by the nature of common soil water constitutive relations
In uniform soils the water content or mixed water content that can lead to solutions with low regularity (Alt and Luckhaus,
forms of Richards’ equation are valuable because water content 1983). Specifically, the solution depends on two empirical,
is a continuous variable. However, in nature soils are seldom highly nonlinear soil water constitutive relations: the unsaturated
uniform over significant length scales, and layered soils are hydraulic conductivity function K(θ) that can be constant or
ubiquitous. In layered soils the water content is discontinuous very near zero for non-positive values of the capillary head, and
across layer interfaces because of unique unsaturated capillary the capillary head function y(θ) that can take on arbitrary small
head relations in the different soil layers (Assouline, 2013). values for relative saturations near 100%. These extremes lead
Rather, the capillary head (y) is continuous, and it is better to to degeneracy in the solution of Richards’ equation. The soil
write the Richards’ equation with capillary head as the dependent constitutive relations may not be smoothly differentiable at these
variable and evaluate the moisture content in terms of y, θ = extremes and can further have very high slopes and hysteresis,
θ(y). This can be done in either a mixed form: and be discontinuous at low relative saturations. Furthermore,
∂ q(y) ∂ infiltration into dry soils often leads to sharp wetting fronts
∂y
− K ( y) −1 =
0 [3] that result in extremely large spatial gradients of soil hydraulic
∂t ∂z ∂z
properties (Zha et al., 2017). These nonlinearities and the
or via the chain rule as the fully head-based form: degeneracy make the design and analysis of numerical schemes for
the Richards’ equation very difficult (Miller et al., 2013).
∂y ∂ ∂ y [4] Exact one-dimensional solutions of Richards’ equation have
c ( y) − K ( y) −1 =
0
∂ t ∂ z ∂z been derived for a few specialized forms of the constitutive relations
where c(y) equals ¶q/¶y equals the specific moisture describing the soil water retention and the unsaturated hydraulic
capacity [L–1]; K(y) equals the hydraulic conductivity conductivity functions (Rogers et al., 1983; Broadbridge and
function written as a function of y[L T–1]. White, 1988; Sander et al., 1988; Barry and Sander, 1991; Barry et
Equations [3] and [4] can both be used to solve unsaturated al., 1993; Ross and Parlange, 1994). However, these solutions are
and saturated flow problems. One may also encounter Eq. [4] in not generally applicable because either the functional forms are
a form that includes fluid compressibility effects dissimilar from widely used constitutive relations that represent
real soils, and/or the solutions impose strict requirements on
www.soils.org/publications/sssaj 1259
the art and identify outstanding issues for the primary numerical an example, Fig. 2 illustrates a typical piecewise linear Lagrange
components of a Richards’ equation solver below. polynomial, wI, that is the building block for most finite element
models for Richards’ equation. It takes a value of 1 at vertex,
Spatial Discretization I, and is zero at the other vertices of the triangulation, Mh. Of
The overwhelming majority of Richards’ equation solvers course, Eq. [6] and [7] are only intended to provide some context
employ either a finite difference, finite volume, or finite element for the following discussion, and we have neglected any details of
approximation in space. Hybrid combinations are also possible the solution approximation or evaluation of the integrals in Eq.
(Helmig, 2011). Table 1 provides a representative list of research [6] and [7] (Kees et al., 2008).
and production codes along with their spatial approximations. The spatial accuracy of all of these approximation methods
The details of these approaches vary. They all build on a discrete depends on several factors. In particular, it is generally a function
partition or tessellation of the spatial domain of interest. The of the local mesh geometry, which can be expressed through a
tessellation may be a structured grid or more generally an characteristic length scale and a measure of the aspect ratio or
unstructured mesh. Figure 1 illustrates a simple unstructured distortion (Trangenstein, 2013). Loosely, the formal accuracy of
mesh of a polygonal spatial domain, W, consisting of a union of a convergent scheme can be related to the characteristic length
non-empty, non-overlapping triangles, Mh = {We}. scale of the mesh raised to some power, which we will refer to as
Each scheme must then approximate the spatial variation the scheme’s order. There is a traditional rule of thumb that large
of material properties and solution variables on this mesh and angles (hence aspect ratios) negatively impact approximation
enforce the conservation statement and solution dynamics accuracy (Shewchuk, 2002). On the other hand, it is also possible
expressed by Richards’ equation in some way. For example, through mesh optimization techniques to exploit solution
conservation might be enforced point-wise in a traditional finite behavior and gain higher accuracy through the appropriate use of
difference scheme. Or, it might be enforced in an integral sense element anisotropy (Pain et al., 2001; Mostaghimi et al., 2015).
using test functions that are constant over a control volume, V, in The spatial discretization size is relevant beyond just the
a finite volume method formal accuracy. Convergence of a method relates to how well
the scheme approximates a solution to Richards’ equation. It
∂q
∫ d x − ∫ ∇⋅[ K (y)(∇ y −1z ) ] d x = 0 [6] does not address whether or not Richards’ equation is a valid
V ∂t V
approximation for unsaturated flow at the length scales used
Here, 1z is a vector denoting the vertical axis. V could correspond in the mesh, however. For example, it is unlikely that a three-
to a mesh element, We, in “cell-centered” scheme or a dual control dimensional solution of Richards’ equation on a large, multiple-
volume, W , in a “vertex-centered” scheme (Fig. 1) (Huber and kilometer-scale voxel can accurately represent nature or be
I
Helmig, 2000; Manzini and Ferraris, 2004). Finite element physically meaningful because it will violate the representative
methods are also weighted residual methods that enforce elementary volume assumption (Or et al., 2015).
Richards’ equation weakly, For reference, Fig. 3 shows three commonly used one-
∂q dimensional discretizations. From left to right these spatial
∫ w d x − ∫ ∇⋅[ K (y)(∇ y −1z ) ] w d =
x 0, ∀ w∈W (W) [7] discretizations represent increasing degrees of complexity, and
W ∂t W
increasing degrees of accuracy. The uniform discretization is
Finite element methods impose a richer structure on the space simplest, but the nature of the solution requires fine resolution
of weighting functions, W(W) (Ern and Guermond, 2004). As everywhere. The variable discretization provides high resolution
near the land surface, where accurately simulating the change
Table 1. Alphabetical list of representative research and pro- in water content with time is most important for accurate
duction Richards’ equation codes. partitioning of rainfall or energy (Downer and Ogden, 2004).
Spatial The variable spatial discretization also allows coarsening
Code approximation Reference away from the land surface, which can reduce computational
ADH CG FEM† Howington et al. 1999
CATHY CG FEM Camporese et al., 2010
FEHM CV FEM Zyvoloski, 2007
FEFLOW CG FEM Trefry and Muffels, 2007
HydroGeoSphere CV FEM Brunner and Simmons, 2012
HYDRUS CG FEM Simunek et al., 2008
ParFlow CCD Kollet and Maxwell, 2006
RichardsFOAM FV Orgogozo et al., 2014
TOUGH IFD Finsterle et al., 2008
VS2D CCD Healy, 2008
WASH123D CG FEM Yeh et al., 2011
†C G FEM, Continuous Galerkin Finite Element Method; CV FEM,
Control Volume Finite Element Method; CCD, Cell-Centered
Differences; FV, Finite Volume; IFD, Integrated Finite Differences. Fig. 1. Example triangulation with dual mesh.
www.soils.org/publications/sssaj 1261
discretizations incorporate the current solution at a time level solving a fixed number of linear systems per time step (Paniconi et
into their approximations and so require solution of at least one al., 1991; Kavetski et al., 2002). Linearly implicit approximations
nonlinear problem at each time step. can be found in either multistep or multistage (e.g., Rosenbrock
To make the following discussion more precise, we assume a schemes) contexts (Farthing et al., 2006). However they are still
method of lines (MOL) approach to eliminate partial derivative much less common than fully implicit techniques.
terms after applying one of the spatial discretizations from Adaptive time discretizations for Richards’ equation are
Section 3. The subsequent discussion would be analogous if much more common than adaptive spatial discretizations. A
we followed a Rothe paradigm and introduced a temporal classical approach is to increase or decrease the time step by a fixed
discretization before the spatial discretization (Lang, 2013). fraction based on the number of iterations taken by the nonlinear
Applying the MOL leads to following semi-discrete problem solver (Paniconi and Putti, 1994). The methods introduced in
when stepping forward from one time level, tn, to the next, tn+1 (Kavetski et al., 2001a, 2001b) and (Tocci et al., 1997) adapt the
time step size and/or order based on estimates of temporal error.
, y , y ′) 0, for t∈(t n , t n+1 ) [9]
F(t= These approaches combine low and high-order approximations
to establish estimates of the local error associated with a solution
Here, yʹÎ Rn is an n-dimensional vector of discrete unknowns using a given time step and approximation order. They accept
or degrees of freedom resulting from the spatial approximation, or reject a solution based on user-selected tolerances and pick a
y¢ is the time derivative for y, and the time step is Δt(n+1). Initial new order and/or time step size using heuristics often motivated
conditions for Eq. [9] are taken from time level n, and we assume by simple control theory (Söderlind, 2006). These approaches
that they are consistent can be orders of magnitude more efficient than fixed time step
or heuristic alternatives, depending on the level of temporal
F(t n , y n , y ′, n ) = 0 [10] accuracy requested (Brenan et al., 1996; Kees and Miller, 2002;
Farthing et al., 2003b). On the other hand, heuristic time
Equation [9] is known as a system of Differential Algebraic stepping based on nonlinear solver performance may outperform
Equations (DAEs) and can be thought of as a generalization of adaptation driven by local truncation error estimates for lower
an Ordinary Differential Equation (ODE) system (Brenan et al., accuracy regimes (D’Haese et al., 2007).
1996). As an example, BDFs are multistep methods that can be It is common to adjust time steps based on nonlinear solver
thought of as introducing an approximation for y¢, convergence as well as temporal truncation error estimates in
mature ODE and DAE solvers (Brenan et al., 1996; Hairer et al.,
y ′, n+1 ≈a n+1 y n+1 + b n [11] 1996). However, traditional approaches are not always well-suited
to highly nonlinear problems, especially when low temporal
and so lead to a discrete nonlinear problem accuracy is acceptable (Farthing et al., 2003b; Söderlind and
Wang, 2006). Efforts to combine nonlinear solver performance
F ( t n+1 , y=
n+1
, y ′, n+1 ) G=
n+1
( y n+1 ) 0 [12] and formal truncation error include Gustafsson and Söderlind
(1997), however, such approaches have not been evaluated
that must be solved to obtain y(n+1). Note that Backward Euler is thoroughly for Richards’ equation and are not common practice.
just the first order BDF
1 1 Linear and Nonlinear Solvers
y ' , n+1 ≈ n+1
y n+1 − n+1 y n [13] The desire to adjust a time step based on the performance of
Dt Dt
the nonlinear solver arises from the fundamentally local nature of
Similarly, an implicit k stage Runge-Kutta approximation for Eq. Newton’s method and its variants that serve as the linearization
[9] requires solution of the nonlinear problems (Brenan et al., techniques for the vast majority of Richards’ equation solvers.
1996; Hairer et al., 1996) That is, returning to the notation for the discrete MOL system
at time level tn+1
( k
)
F t in+1 , y n +D t n+1 ∑ j=1 aij Y ' j , Y ' i =
0, for i =
1,..., k [14]
G n+1 ( y n+1 ) = 0 [16]
for the intermediate stage derivatives Y¢i to obtain
Newton’s method attempts to solve Eq. [16] through a series of
y n+=1 y n + ∑ j=1 b j Y ' j [15]
k
linear problems
Here the implicitness results from one or more coefficient aij >
0, j ≥ i. −G n+1 ( y ni +1 ) , y ni++11 =
J ni +1D y ni +1 = y ni +1 +D y ni +1 , i =
1,... [17]
The nonlinear system of equations in Eq. [12] or [14] must,
∂G
in general, be solved iteratively until some error tolerance is met. until some convergence tolerance is met. Here J ni +1 =
∂ y
( y ni +1 )
On the other hand, there has been interest over the years in linearly is the Jacobian evaluated at y ni+1 Common convergence criteria
implicit or non-iterative temporal approximations, which require include specifying tolerances on the norm of the residual,
www.soils.org/publications/sssaj 1263
(Kornhuber, 2002). While this suffices for homogeneous the proper location of high-resolution space should also improve
domains, additional conditions based on conserving mass robustness and accuracy in many cases.
(via flux matching) and maintaining continuity of pressure at Similarly, many of the issues with robustness of Richards’
material interfaces are required for heterogeneous domains. The equation solvers may be addressed with better understanding
nonlinearity of these transmission conditions also requires a of the dynamic interplay between temporal approximation and
more complex, nonlinear substructuring domain-decomposition linearization techniques. As discussed above, researchers have
approach (Berninger et al., 2014). tried to manage these dynamics through the time step (Tocci
Finally, for large problems with iterative solvers there is et al., 1997; Kavetski et al., 2002; D’Haese et al., 2007) and/or
the need to manage not just the impact of the time step on the linearization technique (Paniconi and Putti, 1994; Lehmann and
difficulty of the nonlinear system, but also its impact on the Ackerer, 1998; Lipnikov et al., 2016). Alternative linearization
resulting linear problems, which are often better conditioned for approaches (e.g., List and Radu, 2016) and better algorithms for
smaller time steps. Moreover, clear efficiency gains are possible managing the complex dynamics of adaptive solvers taken, for
by coordinating the linear and nonlinear solver tolerances example, from control-theory (Bhaya and Kaszkurewicz, 2006)
dynamically to avoid over-solving intermediate linear systems could directly improve existing production codes and also lower
(Tocci et al., 1998; Eisenstat and Walker, 1996). the barrier to adoption of more accurate spatial approximations.
We believe the notions of computational stability and
Discussion and Alternatives proportionality are useful concepts to guide the development
Computational inefficiency and lack of robustness in of modern Richards’ equation solvers. Simulators for Richards’
Richards’ equation solvers used in practice most often manifest equation should have clearly defined input parameters for
as poor linear and nonlinear solver convergence and/or poor controlling their accuracy, and changes in these parameters should
performance of the time integrator. Our view is that a significant reliably increase (or decrease) accuracy and computational expense
portion of this poor convergence and instability arises from at a predictable rate. In other words, tightening an error tolerance
under resolution in space of sharp fronts or other phenomena slightly should lead to a commensurate decrease in error without
where highly nonlinear soil properties change drastically over excessive increases in run time (Söderlind and Wang, 2006).
neighboring elements or cells. However, the spatial component The idea of computational stability follows the line of
of this behavior is not as readily apparent because estimates or introducing mathematical concepts like continuous data
indicators of spatial accuracy are more difficult to obtain than dependence and well-posedness to computational software. The
the convergence estimates and temporal error indicators that are need for more rigorous computational science with standards
available for classical iterative linear and nonlinear techniques like reproducibility and repeatability of numerical experiments
and many time discretizations. Indeed, as discussed above, the has been recognized across a number of fields (Collberg and
vast majority of spatial discretizations used in practice are based Proebsting, 2016). It can be seen, for example, in efforts to
on classical low-order finite element discretizations and cell- develop clear understanding of benchmarking, verification,
centered finite differences and are combined with evaluation and validation for computational codes in fields like nuclear
techniques for soil constitutive relations like mass-lumping and engineering and aerospace engineering (Babuska and Oden,
upwinding that are designed to favor robustness over accuracy. 2004; Oberkampf and Trucano, 2007). A key component of
One way to increase computational efficiency at the such efforts is a suite of test problems identified and accepted
expense of solution validity is to use resolutions considerably by the community that allows fair evaluation and comparison
coarser than the REV scale dictates. This requires the use of so- of numerical techniques and simulation codes. While several
called “effective” parameter values that diverge from physical test problems recur throughout the literature and isolated
reality as the computational scale increases, and results in large comparisons can be found, the community has lacked a central
errors in the simulation’s representation of the physical system repository of benchmark problems or recurring efforts to compare
even though it has converged. We hope instead that there will results across research groups as can be found in other fields
be continued development and adoption of new spatial and (e.g., American Institute of Aeronautics and Astronautics, 1998;
temporal discretization schemes. Following trends of maturing Larson et al., 2014; American Society of Mechanical Engineers,
computational fields, we expect to see improved, adaptive 2017; International Association for the Engineering Modelling,
algorithms driven by solution dynamics and a posteriori error Analysis and Simulation Community, http://www.nafems.
estimates (Mostaghimi et al., 2015; Baron et al., 2017). Moreover, org [verified 14 Sept. 2017]). The recent integrated hydrologic
approaches are needed that consider entire solution error and model comparison project (Maxwell et al., 2014; Kollet et al.,
dynamics–for example, joint space-time discretization and error 2017) and the International Soil Modeling Consortium (http://
control (Solin and Kuraz, 2011) rather than just a MOL approach soil-modeling.org; verified 10 Feb. 2017) represent welcome
with temporal adaptivity but static spatial approximation steps in this direction. This is also an area where government
(Farthing et al., 2003b). While the implementation of full agencies (e.g., NOAA, the US National Weather Service, Office
spatially and temporally adaptive techniques is a clear hurdle, of Water Prediction) could make a major contribution.
www.soils.org/publications/sssaj 1265
saturated groundwater flow. Vadose Zone J. 8(2):352–362. doi:10.2136/
engineering and creation of archival verification and validation
vzj2008.0108
data sets for use by the community are also essential if we are to Bergamaschi, L., B. Luca, and M. Putti. 1999. Mixed finite elements and
proceed rationally and make the most of our efforts. Newton‐type linearizations for the solution of Richards equation. Int.
J. Numer. Methods Eng. 45(8):1025–1046. doi:10.1002/(SICI)1097-
0207(19990720)45:8<1025::AID-NME615>3.0.CO;2-G
Acknowledgments Berninger, H., B. Heiko, K. Ralf, and S. Oliver. 2014. A multidomain
Permission was granted by the Chief of Engineers to publish this discretization of the Richards equation in layered soil. Comput. Geosci.
information. This material is based in part on work supported by 19(1):213–232. doi:10.1007/s10596-014-9461-8
the National Science Foundation under Grant EAR-1360384 to the Berninger, H., R. Kornhuber, and O. Sander. 2011. Fast and robust numerical
University of Wyoming. Any opinions, findings, and conclusions or solution of the Richards equation in homogeneous soil. SIAM J. Numer.
recommendations expressed in this material are those of the author(s) and Anal. 49(6):2576–2597. doi:10.1137/100782887
do not necessarily reflect the views of the National Science Foundation. Bhaya, A., and E. Kaszkurewicz. 2006. Control perspectives on numerical
algorithms and matrix problems. Society for Industrial and Applied
Mathematics, Philadelphia, PA. doi:10.1137/1.9780898718669
References Brenan, K.E., S.L. Campbell, and L.R. Petzold. 1996. Numerical solution
Aavatsmark, I., T. Barkve, O. Bøe, and T. Mannseth. 1998. Discretization of initial-value problems in differential-algebraic equations.
on unstructured grids for inhomogeneous, anisotropic media. Part I: Society for Industrial and Applied Mathematics, Philadelphia, PA.
Derivation of the methods. SIAM J. Sci. Comput. 19(5):1700–1716. doi:10.1137/1.9781611971224
doi:10.1137/S1064827595293582 Broadbridge, P., and I. White. 1988. Modelling solute transport, chemical
Allen, M.B., and C. Murphy. 1985. A finite element collocation method for adsorption and cation exchange. In: Int. Hydrology and Water Resources
variably saturated flows in porous media. Numer. Methods Partial Differ. Symp. Nat. Conf. Publ. No. 92/19 (Preprints of Papers, Vol. 3:924-929).
Equ. 1:229–239. doi:10.1002/num.1690010306 2-4 October. Institution of Engineers, Perth, Australia.
Alt, W.H., and S. Luckhaus. 1983. Quasilinear elliptic-parabolic differential Bronson, S. 2014. Assessing the minimum entropy production rate principle
equations. Math. Z. 183:311–341. doi:10.1007/BF01176474 for multiphase flow using the thermodynamically constrained averaging
American Institute of Aeronautics and Astronautics. 1998. AIAA guide for the theory approach. University of North Carolina, Chapel Hill.
verification and validation of computational fluid dynamics simulations. Brunner, P., and C.T. Simmons. 2012. HydroGeoSphere: A fully integrated,
American Institute of Aeronautics & Astronautics, Reston, VA. physically based hydrological model. Ground Water 50:170–176.
American Society of Mechanical Engineers. 2017. ASME V&V, Verification and doi:10.1111/j.1745-6584.2011.00882.x
validation symposium. American Society of Mechanical Engineers, Las Campbell, J.C., J.M. Hyman, and M.J. Shashkov. 2002. Mimetic finite difference
Vegas, NV. 3–5 May 2017. https://www.asme.org/events/vandv (verified operators for second-order tensors on unstructured grids. Comput. Math.
6 Sept. 2017). Appl. 44(1-2):157–173. doi:10.1016/S0898-1221(02)00137-2
Arbogast, T., A. Todd, C.N. Dawson, P.T. Keenan, M.F. Wheeler, and Y. Ivan. Camporese, M., C. Paniconi, M. Putti and S. Orlandini. 2010. Surface-subsurface
1998. Enhanced cell-centered finite differences for elliptic equations on flow modeling with path-based runoff routing, boundary condition-based
general geometry. SIAM J. Sci. Comput. 19(2):404–425. doi:10.1137/ coupling, and assimilation of multisource observation data. Water Resour.
S1064827594264545 Res. 46. doi:10.1029/2008WR007536.
Arbogast, T., A. Todd, and C. Zhangxin. 1995. On the implementation of mixed Celia, M.A., E.T. Bouloutas, and R.L. Zarba. 1990. A general mass-conservative
methods as nonconforming methods for second-order elliptic problems. numerical solution for the unsaturated flow equation. Water Resour. Res.
Math. Comput. 64(211):943–972. 26(7):1483–1496. doi:10.1029/WR026i007p01483
Arrarás, A., L. Portero, and J.C. Jorge. 2009. A combined mixed finite element Chavent, G., G. Cohen, and J. Jaffre. 1984. Discontinuous upwinding and mixed
ADI scheme for solving Richards’ equation with mixed derivatives on finite elements for two-phase flows in reservoir simulation. Computer
irregular grids. Appl. Numer. Math. 59(3-4):454–467. doi:10.1016/j. Methods in Applied Mechanics and Engineering 47:93–118.
apnum.2008.03.002 Chavent, G., and J.E. Roberts. 1991. A unified physical presentation of mixed,
Assouline, S. 2013. Infiltration into soils: Conceptual approaches and solutions. mixed-hybrid finite elements and standard finite difference approximations
Water Resour. Res. 49:1755–1772. doi:10.1002/wrcr.20155 for the determination of velocities in waterflow problems. Adv. Water
Babuska, I., and J.T. Oden. 2004. Verification and validation in computational Resour. 14(6):329–348. doi:10.1016/0309-1708(91)90020-O
engineering and science: Basic concepts. Comput. Methods Appl. Mech. Chen, Z., G. Huan, and Y. Ma. 2006. Computational methods for multiphase
Eng. 193(36-38):4057–4066. doi:10.1016/j.cma.2004.03.002 flows in porous media. Society for Industrial and Applied Mathematics,
Balay, S., W.D. Gropp, L.C. McInnes, and B.F. Smith. 1997. Efficient management Philadelphia, PA. doi:10.1137/1.9780898718942
of parallelism in object-oriented numerical software libraries. In: E. Arge, Collberg, C., and T.A. Proebsting. 2016. Repeatability in computer systems
A.M. Bruaset, and H.P. Langtengen, editors, Modern software tools for research. Commun. ACM 59(3):62–69.
scientific computing. Springer, New York. p. 163–202. Cox, P.M., R.A. Betts, C.B. Bunton, R.L.H. Essery, P.R. Rowntree, and J. Smith.
Baron, V., Y. Coudière, and P. Sochala. 2017. Adaptive multistep time 1999. The impact of new land surface physics on the GCM simulation
discretization and linearization based on a posteriori error estimates for of climate and climate sensitivity. Clim. Dyn. 15:183–203. doi:10.1007/
the Richards equation. Applied Numerical Mathematics 112:104–125. s003820050276
doi:10.1016/j.apnum.2016.10.005 Dawson, C., S. Sun, and M.F. Wheeler. 2004. Compatible algorithms for
Barry, D.A., and G.C. Sander. 1991. Exact solutions for water infiltration with coupled flow and transport. Comput. Methods Appl. Mech. Eng. 193(23-
an arbitrary surface flux or nonlinear solute adsorption. Water Resour. Res. 26):2565–2580. doi:10.1016/j.cma.2003.12.059
27:2667–2680. doi:10.1029/91WR01445 D’Haese, C.M.F., M. Putti, C. Paniconi, and N.E.C. Verhoest. 2007. Assessment
Barry, D.A., J.-Y. Parlange, G.C. Sander, and M. Sivapalan. 1993. A class of exact of adaptive and heuristic time stepping for variably saturated flow. Int. J.
solutions for Richards’ equation. J. Hydrol. 142:29–46. doi:10.1016/0022- Numer. Methods Fluids 53(7):1173–1193. doi:10.1002/fld.1369
1694(93)90003-R Diersch, H.J. 2013. Flow in variably saturated porous media. In: H.J. Diersch,
Bartlett, R.A. 2009. Integration strategies for computational science and editor, FEFLOW: Finite Element Modeling of Flow, Mass and Heat
engineering software. In: 2009 ICSE Workshop on Software Engineering Transport in Porous and Fractured Media. Springer Verlag. Heidelberg,
for Computational Science and Engineering. Vancouver, BC, Canada. 23- Germany.
23 May 2009. doi:10.1109/SECSE.2009.5069160 Downer, C.W., and F.L. Ogden. 2004. Appropriate vertical discretization of
Bause, M., and P. Knabner. 2004. Computation of variably saturated subsurface Richards’ equation for two-dimensional watershed-scale modelling.
flow by adaptive mixed hybrid finite element methods. Adv. Water Resour. Hydrol. Processes 18:1–22. doi:10.1002/hyp.1306
27(6):565–581. doi:10.1016/j.advwatres.2004.03.005 Eisenstat, S.C., and H.F. Walker. 1996. Choosing the forcing terms in
Belfort, B., B. Benjamin, R. Fanilo, Y. Anis, and L. François. 2009. An an inexact Newton method. SIAM J. Sci. Comput. 17(1):16–32.
efficient lumped mixed hybrid finite element formulation for variably
www.soils.org/publications/sssaj 1267
hydrodynamics: An assessment of the Gothenburg 2010 Workshop. Ogden, F.L., W. Lai, R.C. Steinke, and J. Zhu. 2015b. Validation of finite water-
Springer Science & Business Media, Dordrecht, Germany. content vadose zone dynamics method using column experiments with
Larson, M.G., and A.J. Niklasson. 2004. A conservative flux for the continuous a moving water table and applied surface flux. Water Resour. Res. 51.
Galerkin method based on discontinuous enrichment. Calcolo 41(2):65– doi:10.1002/2014WR016454
76. doi:10.1007/s10092-004-0084-7 Ogden, F.L., M.B. Allen, W. Lai, J. Zhu, M. Seo, C.C. Douglas, and C.A. Talbot.
Lehmann, F., and P.H. Ackerer. 1998. Comparison of iterative methods for 2017. The soil moisture velocity equation. J. Adv. Model. Earth Syst.
improved solutions of the fluid flow equation in partially saturated porous doi:10.1002/2017MS000931
media. Transport in Porous Media 31:275–292. Or, D., P. Lehmann, and S. Assouline. 2015. Natural length scales define the
Li, H., M.W. Farthing, C.N. Dawson, and C.T. Miller. 2007a. Local range of applicability of the Richards equation for capillary flows. Water
discontinuous Galerkin approximations to Richards’ equation. Adv. Water Resour. Res. 51:7130–7144. doi:10.1002/2015WR017034
Resour. 30(3):555–575. doi:10.1016/j.advwatres.2006.04.011 Orgogozo, L., N. Renon, C. Soulaine, F. Henon, S.K. Tomer, D. Labat, et al.
Li, H., M.W. Farthing, and C.T. Miller. 2007b. Adaptive local discontinuous 2014. An open source massively parallel solver for Richards equation:
Galerkin approximation to Richards’ equation. Adv. Water Resour. Mechanistic modelling of water fluxes at the watershed scale. Comput.
30(9):1883–1901. doi:10.1016/j.advwatres.2007.02.007 Phys. Commun. 185:3358–3371. doi:10.1016/j.cpc.2014.08.004
Li, X.S. 2005. An overview of SuperLU. ACM Trans. Math. Softw. 31(3):302– Pain, C.C., A.P. Umpleby, C.R.E. de Oliveira, and A.J.H. Goddard. 2001.
325. doi:10.1145/1089014.1089017 Tetrahedral mesh optimisation and adaptivity for steady-state and
Liang, X., D.P. Lettenmaier, E.F. Wood, and S.J. Burges. 1994. A simple transient finite element calculations. Comput. Methods Appl. Mech. Eng.
hydrologically based model of land surface water and energy fluxes for 190:3771–3796. doi:10.1016/S0045-7825(00)00294-2
general circulation models. J. Geophys. Res. 99(D7):14,415–14,428. Paniconi, C., P. Claudio, A.A. Aldama, and E.F. Wood. 1991. Numerical
doi:10.1029/94JD00483 evaluation of iterative and noniterative methods for the solution of the
Lipnikov, K., L. Konstantin, M. David, and S. Daniil. 2016. New preconditioning nonlinear Richards equation. Water Resour. Res. 27(6):1147–1163.
strategy for Jacobian-free solvers for variably saturated flows with doi:10.1029/91WR00334
Richards’ equation. Adv. Water Resour. 94:11–22. doi:10.1016/j. Paniconi, C., and M. Putti. 1994. A comparison of Picard and Newton
advwatres.2016.04.016 iteration in the numerical solution of multidimensional variably
List, F., and F.A. Radu. 2016. A study on iterative methods for solving Richards’ saturated flow problems. Water Resour. Res. 30(12):3357–3374.
equation. Comput. Geosci. 20:341–353. doi:10.1007/s10596-016-9566-3 doi:10.1029/94WR02046
Lott, P.A., H.F. Walker, C.S. Woodward, and U.M. Yang. 2012. An accelerated Paniconi, C., and M. Putti. 2015. Physically based modeling in catchment
Picard method for nonlinear systems related to variably saturated flow. hydrology at 50: Survey and outlook. Water Resour. Res. 51:7090–7129.
Adv. Water Resour. 38:92–101. doi:10.1016/j.advwatres.2011.12.013 doi:10.1002/2015WR017780
Manzini, G., and S. Ferraris. 2004. Mass-conservative finite volume methods Pettway, J.S., J.H. Schmidt, and A.K. Stagg. 2010. Adaptive meshing in a mixed
on 2-D unstructured grids for the Richards’ equation. Adv. Water Resour. regime hydrologic simulation model. Comput. Geosci. 14(4):665–674.
27(12):1199–1215. doi:10.1016/j.advwatres.2004.08.008 doi:10.1007/s10596-010-9179-1
Maxwell, R.M., M. Putti, S. Meyerhoff, J.-O. Delfs, I.M. Ferguson, V. Ivanov, Philip, J.R. 1957. Numerical solution of equations of the diffusion type
J. Kim, O. Kolditz, S.J. Kollet, M. Kumar, S. Lopez, J. Niu, C. Paniconi, with diffusivity concentration-dependent. II. Aust. J. Phys. 10:29–42.
Y.-J. Park, M.S. Phanikumar, C. Shen, E.A. Sudicky, and M. Sulis. 2014. doi:10.1071/PH570029
Surface-subsurface model intercomparison: A first set of benchmark Pop, I.S., F. Radu, and P. Knabner. 2004. Mixed finite elements for the Richards’
results to diagnose integrated hydrology and feedbacks. Water Resour. Res. equation: Linearization procedure. J. Comput. Appl. Math. 168(1-
50(2):1531–1549. doi:10.1002/2013WR013725 2):365–373. doi:10.1016/j.cam.2003.04.008
McConnell, S. 2004. Code Complete: A practical handbook of software Rannacher, R. 2001. Adaptive Galerkin finite element methods for partial
construction. Microsoft Press, Redmond, WA. differential equations. J. Comput. Appl. Math. 128(1-2):205–233.
Miller, C.T., C. Abhishek, and M.W. Farthing. 2006. A spatially and temporally doi:10.1016/S0377-0427(00)00513-6
adaptive solution of Richards’ equation. Adv. Water Resour. 29(4):525– Rathfelder, K., and L.M. Abriola. 1994. Mass conservative numerical solutions
545. doi:10.1016/j.advwatres.2005.06.008 of the head-based Richards equation. Water Resour. Res. 30:2579–2586.
Miller, C.T., C.N. Dawson, M.W. Farthing, T.Y. Hou, J.F. Huang, C.E. Kees, Richards, L.A. 1931. Capillary conduction of liquids through porous medium.
et al. 2013. Numerical simulation of water resources problems: Models, Physics 1:318–333.
methods, and trends. Adv. Water Resour. 51:405–437. doi:10.1016/j. Richardson, L.F. 1922. Weather prediction by numerical process. University
advwatres.2012.05.008 Press, Cambridge. p. 262.
Miller, C.T., C. George, P.T. Imhoff, J.F. McBride, J.A. Pedit, and J.A. Trangenstein. Rogers, C., M.P. Stallybrass, and D.L. Clements. 1983. On two phase filtration
1998. Multiphase flow and transport modeling in heterogeneous porous under gravity and with boundary infiltration: Application of a Backlund
media: Challenges and approaches. Adv. Water Resour. 21(2):77–120. transformation. Nonlin. Anal. Theory Meth. Appl. 7(7):785–799.
doi:10.1016/S0309-1708(96)00036-X doi:10.1016/0362-546X(83)90034-2
Mostaghimi, P., J.R. Percival, D. Pavlidis, R.J. Ferrier, J.L.M.A. Gomes, G.J. Ross, P.J., and J.-Y. Parlange. 1994. Comparing exact and numerical solutions of
Gorman, et al. 2015. Anisotropic Mesh Adaptivity and Control Volume Richards’ equation for one-dimensional infiltration and drainage. Soil Sci.
Finite Element Methods for Numerical Simulation of Multiphase Flow 157(6):341–344. doi:10.1097/00010694-199406000-00002
in Porous Media. Math. Geosci. 47:417–440. doi:10.1007/s11004-014- Saad, Y., and H.A. van der Vorst. 2000. Iterative solution of linear systems in
9579-1 the 20th century. J. Comput. Appl. Math. 123:1–33. doi:10.1016/S0377-
Niswonger, R.G., and D.E. Prudic. 2009. Comment on “Evaluating interactions 0427(00)00412-X
between groundwater and vadose zone using the HYDRUS-based flow Sander, G.C., J.-Y. Parlange, V. Kuhnel, W.L. Hogarth, D. Lockington, and J.P.J.
package for MODFLOW”. Vadose Zone J. 8:818–819. doi:10.2136/ O’Kane. 1988. Exact nonlinear solution for constant flux infiltration. J.
vzj2008.0155 Hydrol. 97:341–346. doi:10.1016/0022-1694(88)90123-0
Niu, G.-Y., Z.L. Yang, R.E. Dickinson, L.E. Gulden, and H. Su. 2007. Schwab, C. 1998. P- and Hp- finite element methods: Theory and applications in
Development of a simple groundwater model for use in climate models solid and fluid mechanics. Univ. Press, Oxford.
and evaluation with Gravity Recovery and Climate Experiment data. J. Scudeler, C., M. Putti, and C. Paniconi. 2016. Mass-conservative reconstruction
Geophys. Res. 112:D07103. doi:10.1029/2006JD007522 of Galerkin velocity fields for transport simulations. Adv. Water Resour.
Oberkampf, W.L., and T.G. Trucano. 2007. Verification and validation 94:470–485. doi:10.1016/j.advwatres.2016.06.011
benchmarks. Sandia National Laboratory, Albuquerque, NM. Selker, J., and S. John. 2004. Modelling variably saturated flow with HYDRUS-
doi:10.2172/901974 2D. Vadose Zone J. 3(2):725. doi:10.2136/vzj2004.0725
Ogden, F.L., W. Lai, R.C. Steinke, J. Zhu, C.A. Talbot, and J.L. Wilson. 2015a. Shewchuk, J.R. 2002. What is a good linear finite element?- Interpolation,
A new general 1-D vadose zone flow solution method. Water Resour. Res. Conditioning, Anisotropy, and Quality Measures. Proceedings of the 11th
51:4282–4300. doi:10.1002/2015WR017126 International Meshing Roundtable. Sandia National Laboratories and
www.soils.org/publications/sssaj 1269